Voice Authentication Using Multi-Dimensional Acoustic Feature Vectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional voice authentication systems are vulnerable to spoofing attacks, as they rely on voice attributes that can be mimicked by various algorithms, leading to potential unauthorized access.

Innovation Solution

A machine learning multi-dimensional acoustic feature vector authentication system that uses convolutional neural networks (CNNs) to extract and analyze acoustic features, converting them into multi-dimensional vectors for spoofing detection, employing multiple algorithms like Short-Time Fourier Transformation, Mel-Frequency Cepstral Coefficient Transformation, and Tonnetz space geometric transformation to differentiate between human and spoofed voices.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional voice authentication systems use basic voice attribute comparison, then the system is simple to implement, but the system is highly susceptible to spoofing attacks

Engineering Contradiction:
Improveauthentication securityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent transforms voice authentication from traditional time-domain signal comparison to frequency-domain spectral analysis. By converting voice signals to spectrograms and analyzing multi-dimensional acoustic feature vectors in the frequency domain, the system detects spoofing attacks that are invisible in the time domain, thereby improving reliability without excessive complexity increase

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent replaces conventional mechanical voice comparison methods with machine learning-based acoustic feature analysis. Neural networks process multi-dimensional acoustic feature vectors to detect subtle patterns indicating spoofing, substituting simple algorithmic comparison with intelligent pattern recognition while maintaining system feasibility

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If the system uses multiple acoustic feature extraction algorithms to detect spoofing, then the detection accuracy improves, but the processing time and computational complexity increase

Engineering Contradiction:
Improvespoofing detection accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the voice authentication process into distinct stages: acoustic feature extraction, spectral transformation, and spoofing detection. Each stage uses specialized algorithms optimized for its specific task, allowing parallel processing and reducing overall processing time while maintaining high detection accuracy through focused analysis at each stage

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts multiple types of acoustic features (spectral, temporal, and cepstral characteristics) to provide comprehensive spoofing detection. By analyzing more features than the minimum required, the system achieves higher detection accuracy for various spoofing methods, accepting increased computational load as necessary for security-critical applications

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3599606B1Machine learning for authenticating voice
Publication Date: 2023.01.04 ACCENTURE GLOBAL SOLUTIONS LTD
  • EP3599606B1 patent drawingFigure 1
  • EP3599606B1 patent drawingFigure 2
  • EP3599606B1 patent drawingFigure 3

AI summary

A machine learning multi-dimensional acoustic feature vector authentication system, according to an example of the present disclosure, builds and trains multiple multi-dimensional acoustic feature vector machine learning classifiers to determine a probability of spoofing of a voice. The system may extract an acoustic feature from a voice sample of a user. The system may convert the acoustic feature into multi-dimensional acoustic feature vectors and apply the multi-dimensional acoustic feature vectors to the multi-dimensional acoustic feature vector machine learning classifiers to detect spoofing and determine whether to authenticate a user.